Volume 11 Number 9
September 2009
pp. 889–900 889
www.neoplasia.com
Using Optical Spectroscopy to Longitudinally Monitor Physiological Changes within Solid Tumors1
†
‡
Karthik Vishwanath*, Hong Yuan , William T. Barry , § Mark W. Dewhirst and Nimmi Ramanujam* *Department of Biomedical Engineering, Duke University, Durham, NC 27708, USA; †Department of Radiology, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA; ‡Department of Biostatistics and Bioinformatics, Duke University, Durham, NC 27710, USA; §Department of Radiation Oncology, Duke University, Durham, NC 27710, USA
Abstract The feasibility of using quantitative diffuse reflectance spectroscopy to longitudinally monitor physiological response to cancer therapy was evaluated in a preclinical model. This study included two groups of nude mice bearing 4T1 flank tumors (N = 50), half of which were treated with a maximum tolerated dose of doxorubicin (DOX). Diffuse reflectance spectra were collected from tumors during a period of 2 weeks using a fiber-optic probe coupled to a spectrometer. These spectra were quantified using an inverse scalable Monte Carlo model of light transport in tissue to extract the concentrations of oxygenated, deoxygenated hemoglobin (dHb), and a wavelength mean reduced scattering coefficient (). The tumor growth rates of the treated and control groups were nearly identical, as were changes in the scattering parameter during this time frame. However, tumors treated with DOX showed a transient but significant increase in blood oxygen saturation. A comparison between the optically derived and immunohistochemical end points in a subset of the 50 animals showed that the temporal kinetics of dHb concentration and were highly concordant with those of hypoxic and necrotic fractions, respectively. In conclusion, optical methods could function as a “screening” technology in longitudinal studies of small animal tumor models to accelerate development and testing of new anticancer drugs. This technique could isolate specific landmark time points at which more expensive and sophisticated imaging methods or immunohistochemistry could be performed. Neoplasia (2009) 11, 889–900
Introduction Preclinical animal models are widely used in studies seeking to understand cancer biology, particularly, tumor growth, and metastatic behavior [1]. Animal tumor models have provided a template for researchers to understand the genetic basis of molecular pathways involved in tumorigenesis, elucidate the detailed understanding of drug delivery and transport in solid tumors, and investigate the effects of novel therapeutic regimens [2,3]. For decades, the standard method for evaluating the merits of new therapeutic agents in preclinical models has been the standard tumor growth delay assay. This method quantifies the growth rates of control and treated tumors to determine whether a new therapeutic regimen provides any slowing of growth compared with controls. Although this method has been valuable, it hides a plethora of pathologic and physiologic changes that occur in treated tumors that could have important therapeutic implications. A case in point involves “vascular targeting drugs,” such as combrestatin A4. These
compounds cause substantial damage to tumor microvasculature within a few minutes of administration [4] and can lead to complete necrosis of the central portions of a tumor within hours [5]. However, a standard growth delay method exhibits no therapeutic advantage to giving these drugs because the cells at the tumor–normal tissue boundary repopulate Abbreviations: dHb, deoxygenated hemoglobin; DOX, doxorubicin; IHC, immunohistochemistry; MTD, maximum tolerated dose; μ′, s mean reduced scattering; HbO2, oxygenated hemoglobin; MC, Monte Carlo Address all correspondence to: Karthik Vishwanath, Department of Biomedical Engineering, 136 Hudson Hall, Box 90281, Duke University, Durham, NC 27708. E-mail:
[email protected] 1 Funding support for this work was provided by a Department of Defense Era of Hope Scholar award (W81XWH-05-1-0363) and by a National Institutes of Health grant (R01 CA40355-23). Received 6 April 2009; Revised 11 June 2009; Accepted 12 June 2009 Copyright © 2009 Neoplasia Press, Inc. All rights reserved 1522-8002/09/$25.00 DOI 10.1593/neo.09580
890
Longitudinal Optical Sensing of Tumor Physiology
Vishwanath et al.
the tumor very rapidly, aided by a rapid influx of endothelial cell progenitors that promote rapid recovery of angiogenesis [6]. Similarly, it has been shown that inhibition of the hypoxia-inducible factor (HIF-1) causes profound metabolic stress on hypoxic cells because HIF-1 is responsible for upregulating nearly all of the genes involved in anaerobic metabolism [7]. Glucose is the only viable energy substrate for hypoxic cells; thus, in the absence of HIF-1, these cells starve to death, leaving only a small viable rim of tumor cells growing at the margin of the original tumor. Yet, on using a standard growth delay method to evaluate the effects of HIF-1 inhibition, it has been shown that there was virtually no effect on growth delay because growth at the tumor margin hid the underlying changes in tissue physiology and pathology [7]. It is therefore important to assess the underlying tumor physiological and pathologic changes to improve treatment efficacy. One such important parameter that influences tumor growth, metabolism, treatment resistance, and metastatic behavior is tumor hypoxia or low oxygen tension [8,9]. Tissue hypoxia is definitively diagnosed using immunohistochemical assays that directly detect intrinsic markers of hypoxia (such as carbonic anhydrase IX, osteopontin, or HIF-1) or relies on the use of extrinsic agents (such as pimonidazole or EF5) to selectively stain hypoxic cells [10]. A standard nonhistologic method to measure hypoxia is electrode polarography; this yields point measurements of PO2 (sampled from ∼100 to 300 cells) at spatial locations along the path of electrode insertion and withdrawal within the tissue [11]. Tissue oxygen concentrations have also been assessed using perfluorinated compounds in combination with electron paramagnetic resonance spectroscopy, indirectly using 2-nitroimidazoles with positron emission tomography (PET) and magnetic resonance (MR) spectroscopy [10]. Another parameter used to assess the efficacy of cancer therapies involves alterations in tumor morphology and structure including cellular and nuclear features of shape, size, crowding, chromatin organization, and DNA structure [12]. This is most commonly done using histopathologic and immunohistochemistry (IHC) techniques. These methods identify two important features of cell death, namely, apoptosis and necrosis, both of which have the ability to predict disease-free survival in patients after therapy [13–15]. Alternately, PET and MR imaging (MRI) have been applied to noninvasively detect cell death in vivo. MRI measures of the apparent diffusion coefficient of water in tissue can provide information regarding tissue structure from the water content present in the intracellular and the extracellular space, which in turn can be used to assess cell death occurring through necrosis or apoptosis [16,17]. Conversely, PET imaging methods for sensing apoptosis and/or necrosis rely on using labeled apoptosis or necrosis markers such as 18F-labeled annexin V [17]. Studies seeking to quantify changes in physiology including tumor hypoxia and/or cell death using IHC face logistical challenges because these methods require killing animals at regularly spaced intervals to assay them. Moreover, longitudinal studies involving IHC cannot repeatedly track individual subjects because the measurement of an individual using IHC techniques necessitates the loss of the subject. Both these constraints lead to the requirement of large sample sizes to draw meaningful conclusions about a particular therapy. Alternate high-resolution, sophisticated, and noninvasive imaging techniques such as MRI or PET can repeatedly obtain serial measurements on individual subjects, but they have high associated operating costs and may not always be amenable to frequent use in small animal studies. Here, we present an optical spectroscopic approach to nondestructively and dynamically quantify changes in several key biomarkers of carcinogenesis, specifically, tumor oxygenation and cellular morphology
Neoplasia Vol. 11, No. 9, 2009
(necrosis, in particular) in preclinical tumor models. Our technology is based on diffuse reflectance spectroscopy that quantifies the wavelength dependence of the reflected light after it has undergone absorption and scattering interactions in the tissue [18–20]. Primary absorbers in soft tissues include oxygenated hemoglobin (HbO2) and deoxygenated hemoglobin (dHb), both of which indicate blood oxygenation status of the tissue. Tissue scattering contains information about the underlying cellular architecture, is sensitive to the size and density of cellular and subcellular structures such as nuclei and mitochondria, and can be altered by cell death and/or cell proliferation. The sensing depth offered by diffuse reflectance spectroscopy depends on factors that include the source-detector geometry used to deliver and collect the light and the optical properties (absorption and scattering) of the interrogated tissue. Because both the optical absorption and scattering of most tissues decrease with increasing wavelength, the penetration depth of light increases from a few millimeters in the UV-visible spectrum to several centimeters in the near-infrared (NIR). Our technology is designed to be used in the visible spectral range, which provides a sensing depth in the millimeter range and is comparable to the size of tumors in preclinical models such as nude mice. Traditionally, quantitative extraction of absorber concentrations (which yield information on the blood oxygen status and total blood concentration) and tissue scattering are typically obtained using closed-form analytical expressions of light transport in turbid medium, which are best suited for use in the longer, NIR spectral range [21]. We have developed a series of computational Monte Carlo (MC) algorithms that can quantitatively extract tissue absorption and scattering in vivo [22] while accurately accounting for the fiber probe geometry over a wide wavelength range including the visible and NIR. Unlike the closed-form analytical expressions, these computational models are well suited to the visible spectral regime and for small source-detector separations as is proposed for use on tumors in small animal models. This article focuses on the use of visible diffuse reflectance spectroscopy in characterizing changes in optical parameters related to tumor hypoxia and necrosis in vivo in a preclinical model of breast cancer. 4T1 breast tumors were grown in the flank of nude mice, and the animals were separated into control and treatment groups and monitored optically during a 2-week period. The treated animals received a maximum tolerated dose (MTD) of doxorubicin (DOX). These diffuse reflectance spectra were then processed through a quantitative MC model developed by our group [22] to extract the concentrations of HbO2 and dHb in blood within the tumor vasculature along with an associated reduced scattering coefficient spectrum. Tumors from animals in both groups were also resected at four time points for IHC analysis of hypoxic and necrotic fractions. The main findings from our study were as follows: Treatment with MTD DOX showed no differences in the standard growth delays between the treated and untreated tumors. However, tumors treated with DOX showed a transient but significant increase in blood oxygen saturation, which is consistent with previously published studies [23]. A comparison between the optically derived and IHC end points showed that the temporal kinetics of dHb concentration and the reduced scattering coefficient were highly concordant with those of hypoxic and necrotic fractions, respectively. Materials and Methods
Animal Study Protocol Fifty nude mice with a mean ± SD weight of 20.7 ± 3.2 g were ordered from the National Cancer Institute (Bethesda, MD) and stored at
Neoplasia Vol. 11, No. 9, 2009
Longitudinal Optical Sensing of Tumor Physiology
the Duke animal facility in a 12-hour dark-light cycle. Food and water were allowed ad libitum. Approximately, 106 4T1 mouse breast carcinoma cells suspended in 0.1 ml of serum-free medium were inoculated subcutaneously in the right flank of the animals. Once the tumor diameter reached 4 to 6 mm, the animals were evenly distributed (by tumor size) into control and treatment groups (each group had 25 animals). The treatment group received an MTD dose of DOX (10 mg/kg, intravenously using a previously prepared solution of 2-mg/ml DOX), whereas the control group received an equivalent volume of saline. These studies were approved by the Duke Institutional Laboratory Animal Care and Use Committee. The tumors were monitored optically before treatment to get baseline measurements (day 0) and then on days 2, 5, 7, 10, and 13 after DOX or saline administration. The animals were maintained under anesthesia by isoflurane breathing (1.5% isoflurane gas mixed with oxygen) throughout the course of the optical measurements. Five randomly chosen animals, from each group, were removed for IHC on days 0, 5, 10, and 13. These treatment schedules and time points were selected from previous pilot studies that showed maximal differences in the optical end points between the treated and the control groups. The tumors from the animals designated for IHC were resected, snapfrozen by placing the tumor in a container immersed in liquid nitrogen, and stored for subsequent histologic diagnosis and IHC. For each of these animals, the outline of the fiber-optic probe was traced on the surface of the tumor immediately after the optical measurement to ensure that the resected tumors could be sectioned as closely parallel to the face of the fiber-optic probe as possible, thereby yielding sections from nearly the same tissue volumes as sensed by the optical measurements.
Immunohistochemistry The harvested, snap-frozen tumors were sliced into 10-μm-thick sections using a cryotome (CM1850; Leica, Inc., Nussloch, Germany). The orientation of the resected tumor on the microtome before sectioning was determined from the marked circle on the tumor surface that indicated the outline of fiber-optic probe position such that the sections were parallel to the face of the optical probe. For each tumor, five sections were obtained from depths of 600 μm and 1.6 mm below the surface. Slides were counted from the start of cutting and were numbered consecutively so that the depth of each section from the surface could be computed. The hypoxic fraction for each section was determined using pimonidazole and the necrotic fractions were assessed using hematoxylin and eosin (H&E) staining. Pimonidazole as the hypoxia marker was immunostained using a direct labeling protocol: pimonidazole was injected intraperitoneally (at a dose of 60 mg/kg, from a solution of 10-mg/ml pimonidazole) 30 minutes before the animals were killed. The resected tumor sections were fixed in cold acetone for 15 minutes and blocked in donkey serum blocking reagent. The primary antibody to pimonidazole (Hypoxyprobe-1 kits; NPI, Inc, Burlington, MA) was labeled with a fluorescent immunoglobulin G probe (no. Z25005; Molecular Probes, Eugene, OR) and then incubated with tumor sections for 1 hour at room temperature, washed with phosphate-buffered saline, and fixed with 10% buffered formalin for 1 minute. Stained tumor sections were kept in 1% formaldehyde phosphate-buffered saline and imaged within 3 to 4 days. After imaging, the sections were stained with H&E to allow visualization of histologic structure and quantify necrotic fraction.
Quantification of Hypoxic and Necrotic Fractions The hypoxic fraction was determined using pimonidazole-stained images. Theoretically, pimonidazole staining is substantially increased
Vishwanath et al.
891
when the intracellular oxygen tension falls to less than 10 mm Hg [24]. A higher intensity level of the fluorescence staining indicated higher hypoxia in the tissue. Imaged areas whose intensity levels were higher than a given fixed threshold were marked hypoxic, and the hypoxic fraction was calculated as a percentage area of vital tissue with the necrotic areas excluded. Because each resected mass had sections from depths of 600 μm and 1.6 mm, the hypoxic fraction for the tumor was the averaged value from both sections. The necrotic fractions were quantified using H&E-stained sections. The necrotic areas were characterized by higher eosin staining, dark and condensed nuclei debris, and incomplete cell shapes. The necrotic fraction was calculated as the ratio of the necrotic area to the overall tissue area across each section. As before, because there were two sections corresponding to two different depths for each tumor, the averaged values were considered as the necrotic fraction for a given tumor.
Diffuse Reflectance Spectroscopy A fiber-optic–based spectrometer (SkinSkan; JY Horiba, Edison, NJ) was used to measure diffuse reflectance spectra from the tumors. This instrument used a 150-W xenon lamp as the source, its light was filtered through a double-grating excitation monochromator for delivery, and an emission monochromator was coupled to a photomultiplier tube for light detection. The illumination and collection of the light to tissue were achieved through a bifurcated fiber-optic probe (Figure 1B), consisting of a central collection core of 29 fibers (arranged circularly within a diameter of 1.52 mm) surrounded by 30 illumination fibers (arranged nearly concentric to the central core with an outer diameter of 2.18 mm), where each individual fiber was 200 μm in diameter [22]. Diffuse reflectance spectra were measured between 350 and 600 nm by gently pushing the probe to make contact with the tumor surface and held fixed by a clamp throughout the course of the optical measurements. The diffuse reflectance at each wavelength was collected by integrating the signal on the photomultiplier tube for 0.1 second. The acquired spectra were calibrated by dividing the tissue diffuse reflectance spectrum by the diffuse reflectance spectrum obtained from a 99% reflectance Spectralon standard (SRS-99-020; Labsphere, Inc, North Sutton, NH) to account for variations in system throughput and wavelength response of the instrument.
Extraction of HbO2 , dHb, and μ′s Two measurements of diffuse optical reflectance spectra were collected from each animal on each measurement day, consecutively. These reflectance spectra were analyzed using the inverse MC model developed by our group [22] to extract wavelength-dependent absorption and scattering coefficient spectra; here, the tissue absorption/ scattering coefficient is defined as the inverse mean free path for a photon (of given wavelength) to undergo an absorption/scattering event during propagation in the tissue. The tissue absorption coefficient was assumed to have contributions from HbO2, dHb, and skin (extinction coefficients for these absorbers were obtained from a previously tabulated database [25]), whereas the scattering coefficient was calculated using a Mie theory model for spherical scatterers [22]. The following parameters were derived from the absorption and scattering coefficient spectra extracted using the inverse MC model: concentration of HbO2, concentration of dHb, a skin-related absorption factor [26], and the reduced scattering spectrum, which, when averaged, gave a wavelength mean reduced scattering coefficient s [22]. The mean value of each parameter was computed from the two sets of measured diffuse reflectance spectra, for each animal, on each measurement day.
892
Longitudinal Optical Sensing of Tumor Physiology
Vishwanath et al.
Neoplasia Vol. 11, No. 9, 2009
Figure 1. The tissue models, fiber probe schematic, the absorption and scattering coefficients of skin and tumor used in the modeling studies, and the predicted diffuse reflectance spectra. (A) Tissue geometries with and without a skin layer (0.3 mm). (B) Schematic of the composite probe used in these simulations and experiments. (C) Absorption coefficient spectra for the tumor (red crosses) and skin layers (blue crosses). (D) Scattering coefficient spectra for the tumor (red squares) and skin layers (blue squares). (E) Simulated diffuse reflectance spectra for the single-layer model (blue crosses) and the two-layered model (red triangles).
The extracted HbO2 and dHb concentrations were used to compute two additional tissue physiological parameters: the total hemoglobin concentration, [THb] = [HbO2] + [dHb]; and the oxygen saturation, SO2 = 100 × [HbO2] / [THb]. The wavelength range used for data analysis was 480 to 600 nm. This range was selected based on some systematic modeling studies described in the next paragraphs.
The inverse MC model [22] assumes that the interrogated tissue is an optically homogeneous, semi-infinite, turbid medium. However, the experimental measurements are better represented by a two-layered tissue model with a layer of skin covering the underlying tumor, as shown schematically in Figure 1A. To ascertain that we were sensitive to the underlying tumor physiology without being confounded by the presence
Neoplasia Vol. 11, No. 9, 2009
Longitudinal Optical Sensing of Tumor Physiology
of the skin layer, we conducted a series of simulations using standard forward MC simulations [27] for two cases: 1) for a semi-infinite homogeneous tissue model (representing the tumor) and 2) for a two-layered tissue model that included a 300-μm-thick layer of skin atop the semiinfinite tumor, as shown in Figure 1A. These simulations allowed us to estimate perturbations in the diffuse reflectance spectra due to the presence of the skin layer. The skin layer thickness was set to 300 μm in these simulations because the mean ± SD skin thickness in five animals bearing 4T1 tumors was histologically measured to be 290 ± 60 μm. Forward MC simulations were computed for the two tissue models mentioned previously to generate the diffuse reflectance spectrum between 350 and 600 nm at intervals of 10 nm to simulate experimental measurements. It is to be noted that these simulations considered the exact fiber probe geometry shown in Figure 1B for calculations following the procedure outlined previously [22]. Figure 1, C and D, shows representative wavelength-dependent optical absorption and scattering coefficients of each layer, respectively, used in these simulations. The optical properties (absorption and scattering coefficients) of mouse skin were obtained from previously tabulated data [28], whereas those of the tumor were obtained from the results of an inversion procedure from a randomly selected animal in this study. These modeling studies considered the exact geometries of the source-detector fiber pairs of the composite fiber-optic probe to accurately compute the diffuse reflectance for the tissue model, as described elsewhere [22]. Figure 1E shows the simulated diffuse reflectance data for the single-layer (blue crosses) and the two-layered tissue models (red triangles). The presence of the skin layer strongly perturbed the measured diffuse reflectance spectra for wavelengths below 450 nm, whereas for wavelengths above 480 nm, the diffuse reflectance spectra for both models were quite similar. We therefore estimated that the effect of the skin layer on the measured diffuse reflectance was small in the spectral region between 480 and 600 nm and thus used the diffuse reflectance spanning 480 to 600 nm for all extractions of the absorption and scattering coefficients of the tissue.
Statistical Methods To examine the profiles of optical parameters over time, a longitudinal mixed-effects model was applied to these end points as a common method for analyzing repeated-measures data from mouse xenografts models [29,30]. A log transformation was required to stabilize the variance and obtain residuals that were normally distributed for the optical estimates of [dHb], [THb], and s (data not shown). The covariance structure was estimated using a compound symmetric model that was determined to be optimal by the Bayesian information criteria. The significance of fixed effects (i.e., treatment [Controls vs DOX], time point [day], and their interactions) was determined in a step-up fashion using likelihood ratio tests under maximum likelihood estimation. Analyses were performed using R version 2.8.1 (www.r-project.org). For all tests, statistical significance was set at P < .05. Results
Intratumor versus Intertumor Variability in the Extracted Optical End Points To assess variations in each of the extracted optical end points within each animal due to spatial sampling of the tumor relative to the variation in these end points across tumors in different animals, diffuse reflectance spectra were obtained from five different spatial locations on the tumor for each animal of a total of four randomly selected animals (two treated and two control animals) on day 14. The mean coefficients
893
Vishwanath et al.
Table 1. Intertumor versus Intratumor Coefficient of Variation in the Extracted Optical End Points. Optical End Point
HbO2
dHb
s
Mean coefficient of variance within animal (n = 4) Coefficient of variance across animals (n = 10)
0.26 (0.04) 0.62
0.21 (0.11) 0.67
0.12 (0.04) 0.21
Values in parenthesis indicate the SD in the coefficient of variance (computed across the four animals).
of variation for the optical parameters (calculated using the five measurements per animal) within an animal were compared with the coefficient of variation computed for the 10 remaining animals on day 14 (the other 40 animals were previously killed for IHC). These data are tabulated in Table 1 and show that the interanimal coefficient of variation was significantly greater than intra-animal values for all extracted optical parameters.
Representative Optical Absorption and Scattering Spectra Figure 2 shows the diffuse reflectance spectra, the extracted absorption, and reduced scattering spectra for two representative animals (one treated and one control animal) on each measurement day until day 10. Figure 2, A and B, shows the measured diffuse reflectance spectra (symbols) and the corresponding MC model fits (lines) for one representative treated animal (Figure 2A, blue symbols) and one control animal (Figure 2B, red symbols). As seen in these figures, the model fit the measured data well (with mean residual errors